IAC-099 Module 3: AI Editor

Multi-LLM content polishing with privacy-filtered transcripts.

Purpose

Takes privacy-filtered content and provides AI-assisted editing and polishing. Supports multiple LLM providers including fully local Ollama option. ONLY module that requires internet (when using cloud LLMs).

Input/Output Contract

Input: filtered.md - Privacy-filtered markdown from Module 2 Output: polished.md - AI-enhanced content

Tech Stack

• Framework: Tauri 2
• Frontend: Svelte 5
• Backend: FastAPI (local server)
• LLM Providers: Anthropic, OpenAI, OpenRouter, Gemini, Ollama
• Database: SQLite (local storage)
• Platform: macOS (Apple Silicon M2)

Prerequisites

• macOS with Apple Silicon (M2/M3)
• Node.js 20 (via nvm)
• Python 3.11+ with uv
• Rust (latest stable)
• (Optional) Ollama for local AI processing

Installation

# Use Node 20
nvm use 20

# Install frontend dependencies
npm install

# Set up Python backend
cd backend
uv venv
source .venv/bin/activate
uv pip install -r requirements.txt

Development

# Terminal 1: Start FastAPI backend
cd backend
source .venv/bin/activate
uvicorn main:app --reload

# Terminal 2: Start Tauri frontend
npm run tauri:dev

# Run tests
npm test
cd backend && pytest

Building

# Build production app
npm run tauri:build

Usage

1. Load filtered.md from Module 2
2. Select AI provider:
   • Anthropic (Claude) - Requires API key
   • OpenAI (GPT-4) - Requires API key
   • OpenRouter - Requires API key
   • Gemini - Requires API key
   • Ollama - Fully local, no API key needed
3. Click "AI Assist" to get suggestions
4. Review and accept/reject suggestions
5. Click "Export" to save polished.md

LLM Provider Configuration

Cloud Providers (Require Internet + API Keys)

Anthropic (Recommended)

export ANTHROPIC_API_KEY=your-key-here

OpenAI

export OPENAI_API_KEY=your-key-here

OpenRouter

export OPENROUTER_API_KEY=your-key-here

Gemini

export GOOGLE_API_KEY=your-key-here

Local Provider (No Internet Required)

Ollama

# Install Ollama
brew install ollama

# Pull a model
ollama pull llama2

Configuration

• API keys (stored locally, encrypted)
• Default LLM provider
• Model selection per provider
• Temperature/creativity settings
• Max token limits
• System prompts

Source Code References

Based on working implementations from:

• /Users/kjd/01-projects/IAC-033-writer/ - AI editor patterns
• /Users/kjd/01-projects/IAC-011-sovereign-backend/ - API patterns

Success Criteria

• ✅ Loads markdown
• ✅ Connects to at least one LLM (Anthropic or Ollama)
• ✅ Provides AI suggestions
• ✅ Outputs polished markdown
• ✅ Works with API keys (user provides)
• ✅ Ollama option for fully local AI

Architecture

src/
├── components/      # Svelte UI components
│   ├── Editor.svelte
│   └── ProviderSelector.svelte
├── stores/          # Svelte stores
└── tauri/          # Rust backend

backend/
├── main.py         # FastAPI app
├── providers/      # LLM provider integrations
│   ├── anthropic.py
│   ├── openai.py
│   ├── openrouter.py
│   ├── gemini.py
│   └── ollama.py
└── utils/          # Shared utilities

Privacy & Security

• API keys stored locally (encrypted at rest)
• API keys never logged or transmitted except to respective LLM providers
• User explicitly triggers AI assistance (no automatic processing)
• Ollama option provides fully local processing
• No telemetry or analytics

Performance Notes

• Cloud LLMs: Fast response (1-5 seconds)
• Ollama (local): Slower on CPU, faster with GPU (5-30 seconds)
• Recommended Ollama model for M2: llama2 or mistral

License

[To be determined]

Contributing

See main project repository for contribution guidelines.